Abstract

The employment of full-wave electromagnetic (EM) analysis is a practical necessity in the design of contemporary antenna structures. This is because simpler models are generally not available or of limited accuracy. At the same time, EM-based design is computationally expensive. Consequently, the ways of accelerating tasks such as parametric optimization or uncertainty quantification have to be sought. A possible workaround that has been gaining popularity over the recent years is utilization of fast surrogates. Among these, data-driven models are the most popular due to their versatility and easy handling. Notwithstanding, the curse of dimensionality and utility requirements (e.g., having the surrogate valid over broad ranges of geometry and operating parameters) limit the applicability of conventional approximation approaches to antenna modeling. Recently proposed performance-driven methods, especially the nested kriging framework, allow for going beyond the capability of the standard techniques. This is achieved by an appropriate confinement of the model domain, defined to contain only high-quality designs with respect to the selected performance figures. This paper proposes a novel approach, which combines the main idea of the nested kriging, specifically, the first-level model as a tool for constructing the domain-defining manifold, as well as principal component analysis to reduce the domain dimensionality in an explicit manner. Comprehensive benchmarking using three antenna structures indicates superiority of our methodology over conventional techniques, but also nested kriging, in terms of the achievable predictive power. The latter is obtained without compromising design utility of the model as demonstrated through application case studies.

Citations

Authors (2)